>- low performance of ARM hardware vs x86: makes sense considering x86 had been all about performance with no regard to power (until NetBurst forced Intel's hand), and ARM had always been about squeezing long life out of weak batteries

The one thing that kind of bugged me was how he kept referencing ARM as though they made the hardware - they don't - they license it to other companies that then produce it.

>- like many others I thought GC was "solved". It was a real eye-opener to see the RAM requirements to support efficient GC. Makes me wonder what VFP is doing; it's known to run well with what are considered low levels of RAM use (for current desktop hardware), even when working with large tables

I thought VFP used reference counting.

Another interesting read is this one which goes into some of the reasons as to why C# doesn't use reference counting: http://blogs.msdn.com/b/brada/archive/2005/02/11/371015.aspx

I think some of the problems described with cyclic dependencies is a bit overblown - once you understand that it can happen and follow certain patterns for coding, they mostly go away.

One aspect of it that hadn't occurred to me initially was support of multi-threaded code. VFP isn't multi-threaded so it wouldn't have the expense of having to do locks when updating the count. Having said that, even after all of this time I still find .NET's need for IDisposable clunky instead of a more deterministic destructor that you get with ref. counting.

One aspect neglected in the Web perf. article in relation to the performance of GC (Garbage Collection) vs ARC (Automatic Reference Counting) in the graph comparing performance was if there may be a way to optimize GC performance in low memory circumstances. I'd have to believe that the performance of most GCs will be configured for having enough memory (since on desktops that's usually the case). Having said that, I'd expect in cases where having a GC fire is a performance problem you will still need to code in a way which minimizes it (for example, by holding/reusing existing references to keep them from being GC'd), just like the author suggests. But in my mind you'd need to think about it in both cases.